thinkair: the power of mobile cloud computing · 2018. 1. 4. · thinkair: the power of mobile...
TRANSCRIPT
ThinkAir: the power of mobile cloud computingAndrius AucinasIn collaboration with: Sokol Kosta, Pan Hui, Richard Mortier, Xinwen Zhang
What it isThinkAir is a mobile cloud computing framework which allows developers to increase the power of smartphones by using
code offloading. It is built for the Android platform and using VM technology it can bridge mobile devices to any cloud
computing provider to augment the device’s resources.
ThinkAir enables:
• automatic computation offloading from mobile devices to the cloud
• performing on-demand resource allocation, and exploiting parallelism by dynamically creating, resuming, and destroy-
ing VMs when needed.
Example applications
Find all the solutions of the N-Queens Puzzle.
Count the number of faces in a
picture and detect
duplicated photos.
signature 1
signature 2
signature 3
signature N
Return the number of
viruses found.
All solutions to the
N-Queens Puzzle
Number of faces in a picture
and duplicate photos
Count viruses in a filesystem
The ThinkAir framework
Preliminary resultsPhone: app executed entirely on the phone
WiFi Local: code offloaded to "private cloud"
WiFi: code offloaded to public cloud using WiFi
3G: code offloaded to public cloud using 3G
Left to right: Energy consumed by each component on the phone for 8-queens puzzle in
different scenarios, Time taken and Energy consumed on the phone executing 8-queens puzzle
using N = {1, 2, 4, 8} clones.
Left to right: Energy consumed by each component on the phone for Face detection in different
scenarios, Time taken and Energy consumed on the phone executing Face detection using N = {1,
2, 4, 8} clones.
On-demand resource allocation
ThinkAir Framework
app
PhoneC
loud
VirtualizationInternet
- x86
app
ThinkAir Framework
appapp
detectFaces()
app
ThinkAir Framework
app
- x86
app
ThinkAir Framework
app
- x86
app
ThinkAir Framework
app
- x86
app
app
app
appappapp
app
returnresult
Work in progress and further directionsMore accurate resource usage estimation and prediction using symbolic evaluation techniques
Focus on offloading for real-time, interactive applications - low latency requirement
Domain-specific offloading optimisations: image processing, streaming video information ex-
traction (for AR)
Development of simple to use developer interface to the efficient offloading solutions
Related works[1] Eduardo Cuervo, Aruna Balasubramanian, Dae-ki
Cho, Alec Wolman, Stefan Saroiu, Ranveer Chandra,
and Paramvir Bahl. MAUI: making smartphones last
longer with code offload In MobiSys ’10: Proceedings of
the 8th international conference on Mobile systems, applica-
tions, and services. 2010.
[2] Byung-Gon Chun, Sunghwan Ihm, Petros Maniatis,
Mayur Naik, and Ashwin Patti. CloneCloud: Elastic
Execution between Mobile Device and Cloud In Pro-
ceedings of the 6th European Conference on Computer Sys-
tems (EuroSys 2011).